HR
Mike Pheasant
01/05/2017
Data Import
Load raw data (from https://www.kaggle.com/ludobenistant/hr-analytics), add factors, save.
HR <- read_csv("HR_comma_sep.csv", col_types=cols(
satisfaction_level = col_double(),
last_evaluation = col_double(),
number_project = col_integer(),
average_montly_hours = col_integer(),
time_spend_company = col_integer(),
Work_accident = col_integer(),
left = col_integer(),
promotion_last_5years = col_integer(),
sales = col_character(),
salary = col_character()
))
HR <- plyr::rename(HR, replace=c("satisfaction_level"="Satis", "last_evaluation"="Eval",
"number_project"="NumProj", "average_montly_hours"="MonHrs",
"time_spend_company"="Tenure", "Work_accident"="Accdt01",
"left"="Left01", "promotion_last_5years"="Promo5yr01",
"sales"="Dept", "salary"="Salary"))
HR$Salary <- factor(HR$Salary, levels=c("low", "medium", "high"), ordered=TRUE)
HR$Dept <- factor(HR$Dept)
HR$Accdt <- factor(HR$Accdt01 == 1)
HR$Promo5yr <- factor(HR$Promo5yr01 == 1)
HR$Left <- factor(HR$Left01==1)
set.seed(42)
HR$location <- factor(sapply(HR$Dept, location))
write_csv(HR, "HR.csv")Data Summary
Dimensional Summary
summary(HR)## Satis Eval NumProj MonHrs
## Min. :0.0900 Min. :0.3600 Min. :2.000 Min. : 96.0
## 1st Qu.:0.4400 1st Qu.:0.5600 1st Qu.:3.000 1st Qu.:156.0
## Median :0.6400 Median :0.7200 Median :4.000 Median :200.0
## Mean :0.6128 Mean :0.7161 Mean :3.803 Mean :201.1
## 3rd Qu.:0.8200 3rd Qu.:0.8700 3rd Qu.:5.000 3rd Qu.:245.0
## Max. :1.0000 Max. :1.0000 Max. :7.000 Max. :310.0
##
## Tenure Accdt01 Left01 Promo5yr01
## Min. : 2.000 Min. :0.0000 Min. :0.0000 Min. :0.00000
## 1st Qu.: 3.000 1st Qu.:0.0000 1st Qu.:0.0000 1st Qu.:0.00000
## Median : 3.000 Median :0.0000 Median :0.0000 Median :0.00000
## Mean : 3.498 Mean :0.1446 Mean :0.2381 Mean :0.02127
## 3rd Qu.: 4.000 3rd Qu.:0.0000 3rd Qu.:0.0000 3rd Qu.:0.00000
## Max. :10.000 Max. :1.0000 Max. :1.0000 Max. :1.00000
##
## Dept Salary Accdt Promo5yr
## sales :4140 low :7316 FALSE:12830 FALSE:14680
## technical :2720 medium:6446 TRUE : 2169 TRUE : 319
## support :2229 high :1237
## IT :1227
## product_mng: 902
## marketing : 858
## (Other) :2923
## Left location
## FALSE:11428 Brisbane :1801
## TRUE : 3571 London :8925
## San Francisco:4273
##
##
##
## Pairwise Scatter Plots and Distributions for Staying vs Leaving Staff
- Most dimensions show significantly different distributions for Staying vs Leaving staff
- Bimodal distributions for Leaving staff are evident for evaluation, number of projects and monthly hours worked
- Clustering for Leaving staff is evident comparing: satisfaction, evaluation and monthly hours worked; tenure across many dimensions; and between some departments
plot.new()
png(filename="hr-scatter.png", res=300, width = 3000, height = 3000)
my_colors <- brewer.pal(3, "Set2")
samp.sz <- 3750
pct <- round(100*samp.sz/nrow(HR))
data <- HR[sample(nrow(HR), samp.sz),]
scatterplotMatrix(~Satis+Eval+NumProj+MonHrs+Tenure+Accdt01+Promo5yr01+Dept+Salary|Left01, data=data, reg.line="", smoother="", col=my_colors , smoother.args=list(col="grey") , pch=c(3,4), legend.plot=FALSE, main=paste("Scatter Plot Pairs for HR Turnover (sampling ",pct,"% of data)", sep=""))
par(xpd=TRUE, cex=0.7)
Ly <- 1.16 # for inline HTML
Ly <- 1.05 # for PNG
legend(x=0.91, y=Ly, c("Stay", "Leave"), col=my_colors, pch=c(3,4), horiz=TRUE)
dev.off()
Discussion
Pairwise scatterplots and distributions suggest multi-modal distributions and significant clustering between dimensions for the subset of staff that leave. Add features to classify dimensions for:
- evaluation
- satisfaction
- tenure
- number of projects
- monthly hours worked
Pairwise Numeric Correlations
Correlations between all pairs of numeric dimensions are investigated over the full dataset, and within the Leaving and Staying subsets.
Note that three dimensions are binary 0/1 values and need careful interpretation for correlations:
Left01 has the value 1 for leave and 0 for stay
Promo5yr01 has the value 1 for a promotion within the last 5 years and 0 for none
Accdt01 has the value 1 for accident and 0 for none
P-values (and optionally confidence intervals) for all pairs are calculated with the following formula. Correlations not passing the significance test (p < 0.01) will be marked with an ✕ in the matrix.
cor.mtest <- function(mat, conf.level = 0.95){
mat <- as.matrix(mat)
n <- ncol(mat)
p.mat <- lowCI.mat <- uppCI.mat <- matrix(NA, n, n)
diag(p.mat) <- 0
diag(lowCI.mat) <- diag(uppCI.mat) <- 1
for(i in 1:(n-1)){
for(j in (i+1):n){
tmp <- cor.test(mat[,i], mat[,j], conf.level = conf.level)
p.mat[i,j] <- p.mat[j,i] <- tmp$p.value
lowCI.mat[i,j] <- lowCI.mat[j,i] <- tmp$conf.int[1]
uppCI.mat[i,j] <- uppCI.mat[j,i] <- tmp$conf.int[2]
}
}
return(list(p.mat, lowCI.mat, uppCI.mat))
}Overall Correlations (Within the Full Dataset)
Leaving (where 0=stay and 1=leave) is correlated with:
decreasing satisfaction level (-0.39) (negative correlates with decreasing satisfaction)
no work accidents (-0.15) (leaving correlates with fewer work accidents)
no promotion (-0.06) (leaving correlates with no promotion in the last 5 years)
Leaving is positively correlated with:
tenure (0.14) (longer tenure corelates with leaving)
average hours worked per month (0.07) and number of projects worked (0.02)
Other significant correlations include:
satisfaction level is most correlated with last evaluation (0.11), and most negatively with number of projects (-0.14) and tenure (-0.10)
last evaluation is positively correlated with number of projects (0.35), average monthly hours (0.34) and longer tenure (0.13)
tenure correlates with the number of projects (0.20), monthly hours worked (0.13), and receeving a promotion (0.07)
number of projects correlates with monthly hours worked (0.42)
DF <- HR[1:8]
M <- cor(DF)
pval99 <- cor.mtest(DF,0.99)
corrplot(M, p.mat = pval99[[1]], sig.level=0.01, method = 'pie', order ="hclust", addrect=2,
tl.col="black", tl.cex = 1, tl.offset = 0.1, tl.srt = 45)
Discussion
The correlations between leaving and decreasing levels of satisfaction, longer tenure, longer hours of work per month, and low levels of promotion are perhaps not surprising but are important and should be investigated further to try and determine any causes.
The relationship between accidents at work, increasing job satisfaction and lower rates of leaving could also be investigated further as it seems counter-intuitive.
Gather additional data on:
The reasons for leaving
The basis of job satisfaction and tenure
The effectiveness of evaluation procedures
The reasons for (and impact of) persistent over-work or overloading of projects
This data could be gathered through methods including surveys, interviews, exit interviews and project post-mortems.
Correlations for Staying Employees
For staff who stay, the strongest correlation is a negative one between job satisfaction and tenure (-0.17)
Job satisfaction is also negatively correlated with the number of projects worked (-0.09)
Job satisfaction is positively correlated with evaluation (0.09) and monthly hours worked (0.06)
Evaluations are positively correlated with monthly hours worked (0.09), job satisfaction (0.09) and the number of projects worked (0.04)
Tenure is positively correlated with promotion (0.09) and projects worked (0.08)
DF <- HR[HR$Left==FALSE,c(1:6,8)]
M <- cor(DF)
pval99 <- cor.mtest(DF,0.99)
corrplot(M, p.mat = pval99[[1]], sig.level=0.01, method = 'pie', order ="hclust", addrect=2,
tl.col="black", tl.cex = 1, tl.offset = 0.1, tl.srt = 45)
Discussion
Surprisingly, the strongest correlation in the data of staff who stay is decreasing job satisfaction levels with increasing tenure with the company. Unsurprisingly, promotions are correlated with increasing tenure and satisfaction levels are correlated with evaluations.
These staff seem to be staying if they are promoted, happier if they receive good evaluations, but unhappier the longer they stay.
- More information should be gathered about tenure and job satisfaction levels
Correlations for Leaving Employees
For staff who leave:
Satisfaction is negatively correlated with the number of projects (-0.23) and the monthly hours worked (-0.08)
Satisfaction is positively correlated with tenure (0.44) and evaluation (0.18)
Evaluations are positively correlated with monthly hours worked (0.83), the number of projects worked (0.80), tenure (0.78) and job satisfaction (0.18)
Tenure is positively correlated with evaluation (0.78), monthly hours worked (0.66), number of projects worked (0.60) and job satisfaction (0.44)
DF <- HR[HR$Left==TRUE,c(1:6,8)]
M <- cor(DF)
pval99 <- cor.mtest(DF,0.99)
corrplot(M, p.mat = pval99[[1]], sig.level=0.01, method = 'pie', order ="hclust", addrect=2,
tl.col="black", tl.cex = 1, tl.offset = 0.1, tl.srt = 45)
Discussion
For the staff who leave, there are some similarities with those who stay: tenure is also positively correlated with the monthly hours worked and with evaluations.
However, there are some significant differences. Tenure is positively correlated with job satisfaction and is negatively correlated with promotion — the opposite holds for those who stay.
These staff seem to be hard-working, happier the longer they stay and if they are recieving good evaluations, but leaving after not receiving promotions.
- More information should be gathered about tenure and the opportunity for advancement
Exploratory Data Analysis
Career advancement
Many people leaving have been with the company at least 4 years but have not had a promotion in the last 5 years, despite working long hours.
with(HR, coplot(MonHrs ~ jitter(Tenure) |
Promo5yr + Left))
These unpromoted, long-term workers who leave are receiving the highest evaluations.
with(HR, coplot(Eval ~ jitter(Tenure) |
Promo5yr + Left))
These unpromoted, long-term workers who leave are also reporting high levels of satisfaction with the company.
with(HR, coplot(Satis ~ jitter(Tenure) |
Promo5yr + Left))
Of the unpromoted workers putting in long hours who are leaving, some report the lowest levels of satisfaction, but more report very high levels of satisfaction.
The ones who are working the fewest hours report fairly low satisfaction.
with(HR, coplot(MonHrs ~ Satis |
Promo5yr + Left))
The staff on the largest number of projects and who put in the longest hours, with no promotion in 5 years, tend to leave.
with(HR, coplot(MonHrs ~ jitter(NumProj) |
Promo5yr + Left))
Of the unpromoted workers who left, many had both high evaluations and high levels of satisfaction.
with(HR, coplot(Eval ~ Satis |
Promo5yr + Left))
Plot: evaluation, satisfaction | hours + Left evaluation, satisfaction | time with co + Left evaluation, satisfaction | NumProj + Left